Andrew Glassner is a Senior Research Scientist at Weta Digital, where he combines deep learning and computer graphics to help artists produce amazing visual effects for movies and television. Glassner has served as Papers Chair of the SIGGRAPH '94 Papers Committee, Founding Editor of the Journal of Computer Graphics Tools, and Editor-in-Chief of ACM Transactions on Graphics. A popular speaker, Glassner has delivered invited addresses to groups around the world on topics ranging from computer graphics and deep learning to story structure and narrative. Glassner has written and directed live-action and animated films, and was creator-writer-director of an online multiplayer murder-mystery game for The Microsoft Network. He has written novels and screenplays, and is developing a serialized story for podcast.

"Andrew is famous for his ability to teach complex topics that blend mathematics and algorithms, and this work I think is his best yet." Andrew Glassner is a research scientist specializing in computer graphics and deep learning. He is currently a Senior Research Scientist at Weta Digital, where he works on integrating deep learning with the production of world-class visual effects for films and television. He has previously worked as a researcher at labs such as the IBM Watson Lab, Xerox PARC, and Microsoft Research. He was Editor in Chief of ACM TOG, the premier research journal in graphics, and Technical Papers Chair for SIGGRAPH, the premier conference in graphics.

Conditional Value at Risk (CVaR) is a prominent risk measure that is being used extensively in various domains. We develop a new formula for the gradient of the CVaR in the form of a conditional expectation. Based on this formula, we propose a novel sampling-based estimator for the gradient of the CVaR, in the spirit of the likelihood-ratio method. We analyze the bias of the estimator, and prove the convergence of a corresponding stochastic gradient descent algorithm to a local CVaR optimum. Our method allows to consider CVaR optimization in new domains. As an example, we consider a reinforcement learning application, and learn a risk-sensitive controller for the game of Tetris.